Jeffrey Kaye, MDLayton Professor of Neurology & Biomedical EngineeringOregon Center for Aging & TechnologyNIA - Layton Aging & Alzheimer's Disease [email protected]
Continuous in-home assessment: New approaches to assessing health
xx
“This really is an innovative approach, but I’m afraid we can’t consider it. It’s never been done before.”
A fundamental limitation of current dementia research and clinical care... detecting meaningful change
Cardinal features of change - slow decline punctuated with acute, unpredictable events - are challenging to assess with current tools and methods.
Early detection
Detecting meaningful change is hard ... And how to improve detection of change
Baseline 3 years 6 years
Symptoms Reported
Functional range
Change
Is this the disease onset?
Changing the Assessment Paradigm
• New Observations & Discovery• Maximally Effective Clinical Research & Trials• Better Outcomes for Patients & Families
Pervasive Computing Wireless Technologies “Big Data Analytics”
SecureInternet
ComputerActivity
Phone Activity
MedTracker
Pervasive Computing Platform for Assessment: Community-wide ‘Life Lab’Activity, Sleep,
Mobility Time& Location
Doors Opening/Closing
Balance, Body CompositionHeart Rate, Temperature, C02
Kaye et al. Journals of Gerontology: Psychological Sciences, 2011
Device/Sensor “X”
SecureInternet
UNDERSTAND REAL WORLD USELife Lab: Large Scale Deployments Relevant Health & Wellness Measures & Interventions in Everyday Environments
UNDERSTAND THE TECHNOLOGIESPoint of Care ‘Smart Apartment’ Lab:Focused Sensor/Measurement Technology Development & Assessment
UNDERSTAND THE DATAORCATECH Data Repository, Data Aggregation, Measurement Analytics & Outcomes
Research Process & InfrastructureUNDERSTAND THE STAKEHOLDERS/KEY QUESTIONSROI (Response Over Internet) surveys, Focus Groups Participant/End-User Assessment
What can you see?
June - July 2011 June - July 2012June - July 2010
CDR = 0; MMSE = 28 CDR = 0.5; MMSE = 27 CDR = 0.5; MMSE = 28
xx
Differentiation of early MCI: Total Activity & Walking
Hayes et al. Alzheimer's & Dementia, 2008; 4(6): 395-405.
MCINL
Hayes et al. Alzheimers Dement, 2008
MCI cases 9Xmore likely in Slow Group
Trajectories of walking speed over timeDodge, et al. Neurology, 2012
Activity patterns associated with mild cognitive impairment
Differentiation of early MCI: Night-time Behavior & Sleep
Hayes, et al. Alzheimer Dis Assoc Disord. 2014
Normal
NA-MCI
A-MCI
Every Day Cognition: Medication adherence as a measure of cognitive function
• Adherence assessed continuously x 5 wks with MedTracker taking a
• Mean Age - 83 yrs• Based on ADAScog: Lower
Cognition Group vs Higher Cognition Group
Hayes et al., Proceedings : Engineering in Medicine and Biology Soc, 2006; Leen, et al., Technology and Aging, 2007 ; Hayes et al. .Journal of Aging Health, 2009; Hayes et al. Telemedicine Jounal and E-Health, 2009
0
10
20
30
40
50
60
70
80
90
100
Lower Cognition Higher Cognition
% A
dher
ent
Median time within 12.0 mins of goal
Median time within 53.4 mins of goal
Significantly Worse Adherence in Lower Cognition Group
Every Day Cognition: Computer use changes over time in MCI (assessing decline without formal cognitive tests)
• At Baseline: Mean 1.5 hours on computer/per day
• Over time:– Less use days per
month – Less use time
when in session– More variable in
use pattern over time14
16
18
20
0 4 8 12 16 20 24 28 32
Mea
n D
ays
on C
ompu
ter
Months of Continuous Monitoring
Intact MCI
Kaye, et al. Alzheimers Dement. 2014; Silbert et al. submitted, 2015
Active, Frequent Assessments can be Delivered Everyday: RCT to Increase Social Interaction in MCI Using Home-based Technologies
• 6 week RCT of daily 30 min video chats• 89% of all possible sessions completed;
Exceptional adherence – no drop-out• Intervention group improved on
executive/fluency measure.• MCI participants spoke 2985 words on average
while cognitively intact spoke 2423 words during sessions; better discrimination of MCI than conventional tests (animal fluency and delayed list recall)
Dodge et al. Alzheimer's & Dementia: Translational Research & Clinical Interventions, 2015
Direct to home visits: Novel assessment opportunities
http://psych.nyu.edu/freemanlab/research.htm
Alzheimer’s Disease Cooperative Study Home Based Assessment Study (ORCATECH Kiosk System used in HBA Study)
Putting it all together: Pervasive computing ‘Big Data’ for more informative research
Outcome of Interest
Austin, 2015
Putting it all together: High dimensional data fusion model predicting care transitions
Context:Weather, CCI,
living in a retirement
community, etc.
Behavioral -Activity Data: Computer use,
time out of home, etc.
Weekly Self-Report:
Mood, Pain, Falls, ER visits, Visitors, etc…
Annual Clinical Assessment:
Cognition, physical function,
biomarkers, etc.
Demographics:Age, education, socioeconomic
status, etc.
Controls:Number of rooms in
home, etc.
Model CareTransition
Outcome
63,745,978 observations
Austin et al. 2014 GSA
Predicting Care Transitions: Sensitivity Analysis
• Likelihood of a person transitioning within next six months – ROC AUC under curve= 0.974
Austin et al. 2014 GSA 17
Putting it all together: High dimensional data fusion model predicting drug (analgesic) class
Context:Weather, CCI,
living in a retirement
community, etc.
Behavioral -Activity Data: Computer use,
time out of home, etc.
Weekly Self-Report:
Mood, Pain, Falls, ER visits, Visitors, etc…
Annual Clinical Assessment:
Cognition, physical function,
biomarkers, etc.
Demographics:Age, education, socioeconomic
status, etc.
Controls:Number of rooms in
home, etc.
Model Analgesic Class
Outcome
66,172,380 observations
Austin et al. 2015 Under Review
Predicting Drug Class Effects: Case of analgesics
Observation period: July 2011 – March of 2014; 66,172,380 observations
NSAID Opioid
Both
Sensitivity (%) 94.9 65.9 67.4
Specificity (%) 99.9 98.6 99.6
Positive Predictive Value (%)
99.7 82.6 86.1
Negative Predictive Value (%)
99.7 96.6 98.9
CorrectlyClassified (%)
99.6 95.6 98.6
Logistic regression models treated as classifiers (and model fit statistics)
Pervasive Computing Technology in Current Therapeutics Research
• Spectacular progress has been made in applying technology to the basic biology of disease creating an embarrassment of riches of potential treatments.
• In contrast, clinical trials assessment ‘technology’ has not appreciably changed since 1747.
But...• Pervasive computing technologies and selected
biomarkers can radically change the way we conduct clinical research
• This will lead to major advances in detecting prodromal change, managing manifest disease and in transforming the effectiveness of clinical trials.
Harnessing the power of pervasive computing systems: transform the conduct of clinical trials
The Bounty of Biotechnology
The Opportunity of Pervasive Computing
http://www.phrma.org/sites/default/files/Alzheimer%27s%202013.pdf
Challenges to detecting meaningful change in clinical trials
Dodge HH, et al. ADNI Biomarker progressions explain higher variability in stage-specific cognitive decline than baseline values in Alzheimer disease. Alzheimers Dement. 2014.
Improving clinical trials through continuous data collection: Smaller samples, more precise estimates, faster, and ecologically validConventional ApproachGroup Bell curves compared
Courtesy of H. Dodge
Continuously Monitored ApproachIndividual Bell Curves
Distribution can be generated for EACH individual within short duration data accrual periods Walking Speed Observed During the First 90 days
for 2 subjects
Your walking speed ≠ my walking speed OR Your computer use ≠ my computer use
Transforming Clinical Trials with High Frequency, Objective, Continuous Data: “Big Data” for Each Subject
CurrentMethod
ContinuousMeasures
LM Delayed Recall*
ComputerUse**
Walking Speed**
SAMPLE SIZE TO SHOW50% EFFECT
688 10 94
SAMPLE SIZE TO SHOW 40% EFFECT
1076 16 148
SAMPLE SIZE TO SHOW 30% EFFECT
1912 26 262
SAMPLE SIZE TO SHOW 20% EFFECT
4300 58 588
MCI Prevention Trial – Sample Size Estimates• More precise estimates of
the trajectory of change; allows for intra-individualpredictions.
• Reduces required sample size and/or time to identify meaningful change.
• Reduces exposure to harm (fewer needed/ fewer exposed)
• Provides the opportunity to substantially improve efficiency and inform go/no-go decisions of trials.
Dodge, et al. AAIC, 2014
BiomarkerEnrichment
Imaging, CSF, Plasma,Genetics,
other risk factors
Traditional Trial Enrichment:
Stratification accordingto increased risk of AD.
Does not predict who progresses
Population AD PathologyGroup
6 months
Next Generation High Efficiency Clinical Trials (Focus on Phase II, early detection of efficacy)
BehavioralPhenotypeEnrichmentSleep hygiene,
Time out of home,etc…
FurtherEnrichment:
Stratification accordingto who will progress.
Can predict who progresses
Disease Progression Group(s)
AD PathologyGroup
2-3 months
ContinuousAssessmentComputer use, Walking speed,
Activity, Mobility,etc…
Efficient LongitudinalAssessment:
Continuously assessed objective measures.
Detects individual relevantchange rapidly
Disease Progression Group
Continuous Assessment:AD Progression
6 months